Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 12

Question 110

A Knowledge Base returns snippets but the application also needs a narrative. What retrieval mechanism or check should be selected for this case?

Use RetrieveAndGenerate with source attribution, not Retrieve alone.
Call Retrieve only, then represent raw snippets as if the service wrote the narrative.
Use an unrelated Bedrock model response without passing the retrieved evidence or citations.
Insert a citation-looking URL in the final text without verifying retrieved spans.

Explanation

Correct Answer

RetrieveAndGenerate orchestrates retrieval and supported model generation, and can return source citations; Retrieve alone returns source hits.

Incorrect Answers

Retrieve returns evidence; it does not generate a composed answer.

A fluent model reply without the source passages is not a grounded narrative.

A synthetic link is not a provenance-backed citation to the actual Knowledge Base result.

Question 111

An app needs to inspect retrieved chunks before generation. What should be done before treating retrieved content as authoritative evidence?

Call Retrieve and apply custom authorization/factual checks before invoking FM.
Use RetrieveAndGenerate in a fully automatic path before application-specific source inspection.
Treat any retrieved chunk as privileged system instructions before testing its rights.
Only add a more persuasive instruction prompt without exposing hits to validation code.

Explanation

Correct Answer

Retrieve lets the application examine evidence, enforce caller permissions and validate source claims before a separate model invocation.

Incorrect Answers

Combined retrieval/generation removes the requested pre-generation custom authorization checkpoint.

A retrieved document is untrusted data and cannot grant the app permission to show it.

Prompt phrasing cannot replace an explicit evidence and entitlement inspection boundary.

Question 112

A task needs separate SQL facts and policy clauses. Which correction most directly fixes this grounding failure?

Flatten SQL records into prose and use vector similarity alone for exact account balances.
Ask a model to supply policy clauses from memory without retrieving current approved documents.
Run one generic keyword search and assume matching terms cover both database rows and policy exceptions.
Decompose the query into structured and unstructured retrieval steps.

Explanation

Correct Answer

Plan structured SQL lookups and unstructured policy retrieval separately, then reconcile authoritative records and clauses before answering.

Incorrect Answers

Vector similarity does not enforce relational joins or numeric transaction correctness.

Model weights cannot establish the effective policy for this decision.

A single text hit cannot prove both structured facts and the governing unstructured rules.

Question 113

A broad user question contains three independent conditions. What retrieval mechanism or check should be selected for this case?

Retrieve one overall highly similar paragraph and assume it satisfies all three independent conditions.
Apply query decomposition and reconcile evidence per condition.
Omit two conditions because they are expensive to search and answer the easiest one confidently.
Increase generation temperature to encourage the model to fill missing condition evidence.

Explanation

Correct Answer

Query decomposition splits the three requirements into subqueries, then verifies source support for each before synthesizing a combined answer.

Incorrect Answers

An overall similarity score does not prove each separate constraint.

This silently fails the actual composite user request.

Sampling variability cannot replace separately checked supporting sources.

Question 114

A query contains ambiguous abbreviations. What should be done before treating retrieved content as authoritative evidence?

Expand terms with governed synonyms without changing user intent.
Expand every abbreviation to its most common Internet meaning without user/context checks.
Use fuzzy vector similarity as proof that two conflicting acronym expansions are equivalent.
Discard rare original terms from the query after expansion.

Explanation

Correct Answer

Governed synonym expansion improves recall for ambiguous abbreviations only when the user's intended domain and term meaning are preserved.

Incorrect Answers

Common global expansions can be wrong in the user's specialized business domain.

Semantic closeness is not evidence of correct domain-specific terminology.

Removing the user's literal identifier may reduce precision and erase distinguishing constraints.

Question 115

A user asks for an answer not supported by any retrieved source. Which correction most directly fixes this grounding failure?

Attach the URL of a vaguely related document as a citation for an unsupported assertion.
Increase model temperature until a plausible-looking policy paragraph appears.
Return a confident answer based on the model's training corpus while labelling it retrieved.
Abstain or escalate rather than invent a citation.

Explanation

Correct Answer

No retrieved evidence supports the requested claim; abstention or human escalation is correct, not an invented citation or model-only assertion.

Incorrect Answers

A live link does not imply the linked passage entails the claimed fact.

Sampling cannot create authoritative missing source evidence.

This falsifies provenance and misleads the user about evidence coverage.

Question 116

A RAG system retrieves many near-identical paragraphs. What retrieval mechanism or check should be selected for this case?

Deduplicate passages and set a diversity-aware context budget.
Increase top-k but retain every near-identical passage unchanged.
Use a longer response maximum while keeping the redundant retrieved context.
Delete the entire duplicate source document from the authoritative source-of-record.

Explanation

Correct Answer

Deduplicate overlapping passages or diversify retrieved chunks so the finite context budget covers distinct evidence rather than repetitions.

Incorrect Answers

More repeats consume context without adding substantive new evidence.

Output capacity does not make repetitive input sources more diverse.

Repeated retrieval hits should be controlled without destroying legitimate source records.

Question 117

A KB uses hybrid search with a store lacking filterable text. What should be done before treating retrieved content as authoritative evidence?

Force HYBRID on an unsupported vector store and assume keyword indexing is automatic.
Rename the vector field but never add a filterable text representation.
Check vector store support and fallback to semantic search when hybrid is unsupported.
Return only the exact lexical match and discard semantic candidates when hybrid is unavailable.

Explanation

Correct Answer

HYBRID is conditional on supported stores and a filterable text field; use SEMANTIC or redesign the store when those prerequisites are absent.

Incorrect Answers

The feature depends on the index/store's text-field support, not the request flag alone.

A field label does not create full-text search capability.

Pure lexical matching may miss relevant paraphrases; select a supported fallback and evaluate recall.

Question 118

A search request returns restricted documents through model-tool access. Which correction most directly fixes this grounding failure?

Filter only the natural-language prompt while exposing every retrieved private document to the tool.
Give the search tool a shared administrator role and rely on the model to hide sensitive passages.
Enforce caller-specific document permissions before tool result publication.
Strip unauthorized citation URLs after sending the private document text to the model.

Explanation

Correct Answer

Authorize every document hit against the verified caller before it enters the model or tool response; tool possession alone grants no source access.

Incorrect Answers

Prompt filtering does not enforce row/document-level entitlements on tool results.

Model self-restraint is not a caller-specific access-control boundary.

Removing links does not undo the disclosure of restricted source content.

Question 119

A model asks for retrieval via MCP tools. What retrieval mechanism or check should be selected for this case?

Execute any discovered MCP tool just because its name contains search.
Validate MCP tool schema and identity access for that search API.
Allow arbitrary tool arguments by ignoring the advertised input schema.
Send a shared privileged token with every MCP call regardless of user scope.

Explanation

Correct Answer

Check MCP server identity, permitted tool capabilities and schema/argument validation before an authenticated search operation.

Incorrect Answers

Discovery does not establish trust, authorization, or argument safety.

Invalid or broad arguments can bypass intended search restrictions.

Shared admin identity does not implement per-caller data authorization.

Question 120

A KB citation links to an inaccessible internal bucket. What should be done before treating retrieved content as authoritative evidence?

Expose s3://private-bucket/key in the answer and claim it is clickable evidence.
Make the entire bucket public so the assistant's citations always work.
Provide authorized citation links, not raw private objects to end users.
Replace the source link with a fake public URL without checking what it serves.

Explanation

Correct Answer

A citation must use an authorized reader path, such as a checked application route, not reveal an inaccessible private S3 object URI.

Incorrect Answers

A raw private object locator is neither a user-authorized URL nor a readable citation.

This leaks unrelated private documents and defeats object-level access controls.

Invented links do not provide traceable evidence and may misrepresent content access.

Pass Your Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Exam Easy!

Question 110

Explanation

Correct Answer

Incorrect Answers

Question 111

Explanation

Correct Answer

Incorrect Answers

Question 112

Explanation

Correct Answer

Incorrect Answers

Question 113

Explanation

Correct Answer

Incorrect Answers

Question 114

Explanation

Correct Answer

Incorrect Answers

Question 115

Explanation

Correct Answer

Incorrect Answers

Question 116

Explanation

Correct Answer

Incorrect Answers

Question 117

Explanation

Correct Answer

Incorrect Answers

Question 118

Explanation

Correct Answer

Incorrect Answers

Question 119

Explanation

Correct Answer

Incorrect Answers

Question 120

Explanation

Correct Answer

Incorrect Answers

AWS Certified Generative AI Developer - Professional AIP-C01 Practice Tests